def get_parser():
parser = argparse.ArgumentParser()
subparsers = parser.add_subparsers(title='Models trainer script',
description='available models')
for model in MODELS.get_model_names():
add_train_args(
MODELS.get_model_train_parser(model)(subparsers.add_parser(model)))
return parser
def get_parser():
parser = add_train_args(organ_parser())
parser.add_argument(
'--n_ref_subsample',
type=int,
default=500,
help='Number of reference molecules (sampling from training data)')
parser.add_argument('--addition_rewards',
nargs='+',
type=str,
choices=MetricsReward.supported_metrics,
default=[],
help='Adding of addition rewards')
# conditional generation
parser.add_argument('--conditional',
type=int,
default=0,
help='Conditional generation mode')
parser.add_argument('--output_size',
type=int,
default=10,
help='Output size in the condition linear layer')
return parser
def get_parser():
parser = add_train_args(vae_parser())
# conditional generation
parser.add_argument('--conditional', type=int, default=0,
help='Conditional generation mode')
parser.add_argument('--output_size', type=int, default=10,
help='Output size in the condition linear layer')
return parser
def get_parser():
parser = add_train_args(organ_parser())
parser.add_argument(
'--n_ref_subsample',
type=int,
default=500,
help='Number of reference molecules (sampling from training data)')
parser.add_argument('--addition_rewards',
nargs='+',
type=str,
choices=MetricsReward.supported_metrics,
default=[],
help='Adding of addition rewards')
return parser
def get_parser():
return add_train_args(vae_parser())
def get_parser():
parser = argparse.ArgumentParser()
subparsers = parser.add_subparsers(
title='Models trainer script', description='available models'
)
for model in MODELS.get_model_names():
add_train_args(
MODELS.get_model_train_parser(model)(
subparsers.add_parser(model)
)
)
parser.add_argument('--model', type=str, default='all',
choices=['all'] + MODELS.get_model_names(),
help='Which model to run')
parser.add_argument('--test_path',
type=str, required=False,
help='Path to test molecules csv')
parser.add_argument('--test_scaffolds_path',
type=str, required=False,
help='Path to scaffold test molecules csv')
parser.add_argument('--train_path',
type=str, required=False,
help='Path to train molecules csv')
parser.add_argument('--ptest_path',
type=str, required=False,
help='Path to precalculated test npz')
parser.add_argument('--ptest_scaffolds_path',
type=str, required=False,
help='Path to precalculated scaffold test npz')
parser.add_argument('--checkpoint_dir', type=str, default='./checkpoints',
help='Directory for checkpoints')
parser.add_argument('--n_samples', type=int, default=30000,
help='Number of samples to sample')
parser.add_argument('--n_jobs', type=int, default=1,
help='Number of threads')
parser.add_argument('--device', type=str, default='cpu',
help='GPU device index in form `cuda:N` (or `cpu`)')
parser.add_argument('--metrics', type=str, default='metrics.csv',
help='Path to output file with metrics')
parser.add_argument('--train_size', type=int, default=None,
help='Size of training dataset')
parser.add_argument('--test_size', type=int, default=None,
help='Size of testing dataset')
parser.add_argument('--experiment_suff', type=str, default='',
help='Experiment suffix to break ambiguity')
# Model
model_arg = parser.add_argument_group('Model')
model_arg.add_argument('--q_cell',
type=str, default='gru', choices=['gru'],
help='Encoder rnn cell type')
model_arg.add_argument('--q_bidir',
default=False, action='store_true',
help='If to add second direction to encoder')
model_arg.add_argument('--q_d_h',
type=int, default=256,
help='Encoder h dimensionality')
model_arg.add_argument('--q_n_layers',
type=int, default=1,
help='Encoder number of layers')
model_arg.add_argument('--q_dropout',
type=float, default=0.5,
help='Encoder layers dropout')
model_arg.add_argument('--d_cell',
type=str, default='gru', choices=['gru'],
help='Decoder rnn cell type')
model_arg.add_argument('--d_n_layers',
type=int, default=3,
help='Decoder number of layers')
model_arg.add_argument('--d_dropout',
type=float, default=0,
help='Decoder layers dropout')
model_arg.add_argument('--d_z',
type=int, default=128,
help='Latent vector dimensionality')
model_arg.add_argument('--d_d_h',
type=int, default=512,
help='Decoder hidden dimensionality')
model_arg.add_argument('--freeze_embeddings',
default=False, action='store_true',
help='If to freeze embeddings while training')
# Train
train_arg = parser.add_argument_group('Train')
train_arg.add_argument('--n_batch',
type=int, default=512,
help='Batch size')
train_arg.add_argument('--clip_grad',
type=int, default=50,
help='Clip gradients to this value')
train_arg.add_argument('--kl_start',
type=int, default=0,
help='Epoch to start change kl weight from')
train_arg.add_argument('--kl_w_start',
type=float, default=0,
help='Initial kl weight value')
train_arg.add_argument('--kl_w_end',
type=float, default=0.05,
help='Maximum kl weight value')
train_arg.add_argument('--lr_start',
type=float, default=3 * 1e-4,
help='Initial lr value')
train_arg.add_argument('--lr_n_period',
type=int, default=10,
help='Epochs before first restart in SGDR')
train_arg.add_argument('--lr_n_restarts',
type=int, default=10,
help='Number of restarts in SGDR')
train_arg.add_argument('--lr_n_mult',
type=int, default=1,
help='Mult coefficient after restart in SGDR')
train_arg.add_argument('--lr_end',
type=float, default=3 * 1e-4,
help='Maximum lr weight value')
train_arg.add_argument('--n_last',
type=int, default=1000,
help='Number of iters to smooth loss calc')
train_arg.add_argument('--n_workers',
type=int, default=1,
help='Number of workers for DataLoaders')
return parser
def get_parser():
return add_train_args(char_rnn_parser())
def get_parser():
return add_train_args(junction_tree_parser())
